Glossary
34terms across 10 crates. Each links back to where it's introduced.
A
- AGI (Artificial General Intelligence)
- Hypothetical AI that can learn and perform any intellectual task a human can. Does not exist yet.→ Crate #10
- AI Agent
- An AI system that can take autonomous actions in the world using tools, not just generate text. Follows an observe-think-act loop.→ Crate #10
- Artificial Intelligence (AI)
- A human-made system that can learn, reason, and solve problems. Not one single invention — more like a toolbox of techniques.→ Crate #1
- Attention Mechanism
- The key innovation in Transformers — lets the model learn which words (or tokens) to focus on when processing each part of the input.→ Crate #6
B
- Backpropagation
- The algorithm for training neural networks — calculates error, works backwards through layers, and adjusts weights to reduce future errors.→ Crate #4
- Bias (in data)
- When training data over-represents or under-represents certain groups, leading the model to perform unevenly.→ Crate #3
C
- CNN (Convolutional Neural Network)
- A neural network designed for images. Slides small filters across the image to detect patterns at increasing levels of complexity.→ Crate #5
- Computer Vision
- The field of making computers understand images and video — classification, detection, segmentation, and generation.→ Crate #5
D
- Deep Learning
- Neural networks with many hidden layers. 'Deep' refers to the layer count, not philosophical depth.→ Crate #4
- Deepfake
- AI-generated fake video or audio depicting real people saying or doing things they never did.→ Crate #8
- Diffusion Model
- A generative model that learns to remove noise from images. Generation starts with random noise and progressively denoises it into a coherent image.→ Crate #9
G
- GAN (Generative Adversarial Network)
- Two networks competing: a generator creates fakes, a discriminator detects them. They improve by training against each other.→ Crate #9
H
- Hallucination
- When an AI model confidently generates information that sounds plausible but is factually incorrect.→ Crate #1
L
- Learning Rate
- How much weights are adjusted on each training step. Too high = overshooting. Too low = painfully slow convergence.→ Crate #4
- LLM (Large Language Model)
- A massive transformer trained on internet-scale text to predict the next word. At scale, this produces surprisingly general capabilities.→ Crate #6
M
- Machine Learning (ML)
- A subset of AI where computers learn rules from data rather than being explicitly programmed.→ Crate #2
N
- Narrow AI
- AI that excels at one specific task (like playing chess or recognizing faces) but can't generalize to other tasks. All current AI is narrow.→ Crate #1
- Neural Network
- A model architecture inspired by biological neurons. Layers of simple mathematical units that collectively learn complex patterns.→ Crate #4
- NLP (Natural Language Processing)
- The field of making computers understand, generate, and work with human language.→ Crate #6
O
- Object Detection
- Identifying what objects are in an image AND where each one is located (bounding boxes).→ Crate #5
- Overfitting
- When a model memorizes training data (including noise) instead of learning general patterns. Performs well on training data, poorly on new data.→ Crate #7
P
- Parameters / Weights
- The millions (or billions) of numbers inside a model that get adjusted during training. They encode what the model has learned.→ Crate #2
R
- ReAct Pattern
- Reason + Act — the core agent loop: observe the situation, reason about what to do (using an LLM), take an action (using a tool), observe the result, repeat.→ Crate #10
- Reinforcement Learning
- Learning by trial and error. The agent takes actions, receives rewards or penalties, and adjusts its strategy.→ Crate #2
S
- Supervised Learning
- Training a model with both inputs (questions) and correct outputs (answers). The most common form of ML.→ Crate #2
- Synthetic Data
- Artificially generated data that mimics real data. Used to augment limited real datasets.→ Crate #3
T
- Temperature
- A setting that controls randomness in AI output. Higher temperature = more creative/random. Lower = more deterministic/focused.→ Crate #6
- Tool Use
- Giving an AI access to external tools (browser, code runner, APIs) so it can do real work beyond generating text.→ Crate #10
- Training Data
- The examples fed to a model during training. Quality and diversity of this data directly determine model performance.→ Crate #3
- Transformer
- The dominant neural network architecture for language (and increasingly other domains). Uses 'attention' to learn which parts of the input are relevant to each other.→ Crate #6
- Turing Test
- A test proposed by Alan Turing: if a human can't tell whether they're chatting with a machine or a person, the machine passes.→ Crate #1
U
- Underfitting
- When a model is too simple to capture the patterns in the data.→ Crate #7
- Unsupervised Learning
- Training a model with data but no labels. The model discovers patterns and groups on its own.→ Crate #2
W
- Word Embeddings
- Representing words as lists of numbers (vectors) where similar words have similar vectors. Enables math on language: King - Man + Woman = Queen.→ Crate #6